src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSw_VdwNone_GeomP1P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_256_single.h"
  50 #include "kernelutil_x86_avx_256_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_avx_256_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Particle-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_VF_avx_256_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrE,jnrF,jnrG,jnrH;
  78     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  79     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  80     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  81     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  82     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  83     real             rcutoff_scalar;
  84     real             *shiftvec,*fshift,*x,*f;
  85     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  86     real             scratch[4*DIM];
  87     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  88     real *           vdwioffsetptr0;
  89     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  91     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m256i          ewitab;
  96     __m128i          ewitab_lo,ewitab_hi;
  97     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  99     real             *ewtab;
 100     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 101     real             rswitch_scalar,d_scalar;
 102     __m256           dummy_mask,cutoff_mask;
 103     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 104     __m256           one     = _mm256_set1_ps(1.0);
 105     __m256           two     = _mm256_set1_ps(2.0);
 106     x                = xx[0];
 107     f                = ff[0];
 108
 109     nri              = nlist->nri;
 110     iinr             = nlist->iinr;
 111     jindex           = nlist->jindex;
 112     jjnr             = nlist->jjnr;
 113     shiftidx         = nlist->shift;
 114     gid              = nlist->gid;
 115     shiftvec         = fr->shift_vec[0];
 116     fshift           = fr->fshift[0];
 117     facel            = _mm256_set1_ps(fr->epsfac);
 118     charge           = mdatoms->chargeA;
 119
 120     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 121     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 122     beta2            = _mm256_mul_ps(beta,beta);
 123     beta3            = _mm256_mul_ps(beta,beta2);
 124
 125     ewtab            = fr->ic->tabq_coul_FDV0;
 126     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 127     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 128
 129     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 130     rcutoff_scalar   = fr->rcoulomb;
 131     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 132     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 133
 134     rswitch_scalar   = fr->rcoulomb_switch;
 135     rswitch          = _mm256_set1_ps(rswitch_scalar);
 136     /* Setup switch parameters */
 137     d_scalar         = rcutoff_scalar-rswitch_scalar;
 138     d                = _mm256_set1_ps(d_scalar);
 139     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 140     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 141     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 142     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 143     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 144     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150     j_coord_offsetC = 0;
 151     j_coord_offsetD = 0;
 152     j_coord_offsetE = 0;
 153     j_coord_offsetF = 0;
 154     j_coord_offsetG = 0;
 155     j_coord_offsetH = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     for(iidx=0;iidx<4*DIM;iidx++)
 161     {
 162         scratch[iidx] = 0.0;
 163     }
 164
 165     /* Start outer loop over neighborlists */
 166     for(iidx=0; iidx<nri; iidx++)
 167     {
 168         /* Load shift vector for this list */
 169         i_shift_offset   = DIM*shiftidx[iidx];
 170
 171         /* Load limits for loop over neighbors */
 172         j_index_start    = jindex[iidx];
 173         j_index_end      = jindex[iidx+1];
 174
 175         /* Get outer coordinate index */
 176         inr              = iinr[iidx];
 177         i_coord_offset   = DIM*inr;
 178
 179         /* Load i particle coords and add shift vector */
 180         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 181
 182         fix0             = _mm256_setzero_ps();
 183         fiy0             = _mm256_setzero_ps();
 184         fiz0             = _mm256_setzero_ps();
 185
 186         /* Load parameters for i particles */
 187         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 188
 189         /* Reset potential sums */
 190         velecsum         = _mm256_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             jnrE             = jjnr[jidx+4];
 202             jnrF             = jjnr[jidx+5];
 203             jnrG             = jjnr[jidx+6];
 204             jnrH             = jjnr[jidx+7];
 205             j_coord_offsetA  = DIM*jnrA;
 206             j_coord_offsetB  = DIM*jnrB;
 207             j_coord_offsetC  = DIM*jnrC;
 208             j_coord_offsetD  = DIM*jnrD;
 209             j_coord_offsetE  = DIM*jnrE;
 210             j_coord_offsetF  = DIM*jnrF;
 211             j_coord_offsetG  = DIM*jnrG;
 212             j_coord_offsetH  = DIM*jnrH;
 213
 214             /* load j atom coordinates */
 215             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 216                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 217                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 218                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 219                                                  &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm256_sub_ps(ix0,jx0);
 223             dy00             = _mm256_sub_ps(iy0,jy0);
 224             dz00             = _mm256_sub_ps(iz0,jz0);
 225
 226             /* Calculate squared distance and things based on it */
 227             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 228
 229             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 230
 231             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 232
 233             /* Load parameters for j particles */
 234             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 235                                                                  charge+jnrC+0,charge+jnrD+0,
 236                                                                  charge+jnrE+0,charge+jnrF+0,
 237                                                                  charge+jnrG+0,charge+jnrH+0);
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 244             {
 245
 246             r00              = _mm256_mul_ps(rsq00,rinv00);
 247
 248             /* Compute parameters for interactions between i and j atoms */
 249             qq00             = _mm256_mul_ps(iq0,jq0);
 250
 251             /* EWALD ELECTROSTATICS */
 252
 253             /* Analytical PME correction */
 254             zeta2            = _mm256_mul_ps(beta2,rsq00);
 255             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 256             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 257             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 258             felec            = _mm256_mul_ps(qq00,felec);
 259             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 260             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 261             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 262             velec            = _mm256_mul_ps(qq00,velec);
 263
 264             d                = _mm256_sub_ps(r00,rswitch);
 265             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 266             d2               = _mm256_mul_ps(d,d);
 267             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 268
 269             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 270
 271             /* Evaluate switch function */
 272             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 273             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 274             velec            = _mm256_mul_ps(velec,sw);
 275             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             velec            = _mm256_and_ps(velec,cutoff_mask);
 279             velecsum         = _mm256_add_ps(velecsum,velec);
 280
 281             fscal            = felec;
 282
 283             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 284
 285             /* Calculate temporary vectorial force */
 286             tx               = _mm256_mul_ps(fscal,dx00);
 287             ty               = _mm256_mul_ps(fscal,dy00);
 288             tz               = _mm256_mul_ps(fscal,dz00);
 289
 290             /* Update vectorial force */
 291             fix0             = _mm256_add_ps(fix0,tx);
 292             fiy0             = _mm256_add_ps(fiy0,ty);
 293             fiz0             = _mm256_add_ps(fiz0,tz);
 294
 295             fjptrA             = f+j_coord_offsetA;
 296             fjptrB             = f+j_coord_offsetB;
 297             fjptrC             = f+j_coord_offsetC;
 298             fjptrD             = f+j_coord_offsetD;
 299             fjptrE             = f+j_coord_offsetE;
 300             fjptrF             = f+j_coord_offsetF;
 301             fjptrG             = f+j_coord_offsetG;
 302             fjptrH             = f+j_coord_offsetH;
 303             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 304
 305             }
 306
 307             /* Inner loop uses 108 flops */
 308         }
 309
 310         if(jidx<j_index_end)
 311         {
 312
 313             /* Get j neighbor index, and coordinate index */
 314             jnrlistA         = jjnr[jidx];
 315             jnrlistB         = jjnr[jidx+1];
 316             jnrlistC         = jjnr[jidx+2];
 317             jnrlistD         = jjnr[jidx+3];
 318             jnrlistE         = jjnr[jidx+4];
 319             jnrlistF         = jjnr[jidx+5];
 320             jnrlistG         = jjnr[jidx+6];
 321             jnrlistH         = jjnr[jidx+7];
 322             /* Sign of each element will be negative for non-real atoms.
 323              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 324              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 325              */
 326             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 327                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 328
 329             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 330             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 331             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 332             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 333             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 334             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 335             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 336             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 337             j_coord_offsetA  = DIM*jnrA;
 338             j_coord_offsetB  = DIM*jnrB;
 339             j_coord_offsetC  = DIM*jnrC;
 340             j_coord_offsetD  = DIM*jnrD;
 341             j_coord_offsetE  = DIM*jnrE;
 342             j_coord_offsetF  = DIM*jnrF;
 343             j_coord_offsetG  = DIM*jnrG;
 344             j_coord_offsetH  = DIM*jnrH;
 345
 346             /* load j atom coordinates */
 347             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 348                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 349                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 350                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 351                                                  &jx0,&jy0,&jz0);
 352
 353             /* Calculate displacement vector */
 354             dx00             = _mm256_sub_ps(ix0,jx0);
 355             dy00             = _mm256_sub_ps(iy0,jy0);
 356             dz00             = _mm256_sub_ps(iz0,jz0);
 357
 358             /* Calculate squared distance and things based on it */
 359             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 360
 361             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 362
 363             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 364
 365             /* Load parameters for j particles */
 366             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 367                                                                  charge+jnrC+0,charge+jnrD+0,
 368                                                                  charge+jnrE+0,charge+jnrF+0,
 369                                                                  charge+jnrG+0,charge+jnrH+0);
 370
 371             /**************************
 372              * CALCULATE INTERACTIONS *
 373              **************************/
 374
 375             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 376             {
 377
 378             r00              = _mm256_mul_ps(rsq00,rinv00);
 379             r00              = _mm256_andnot_ps(dummy_mask,r00);
 380
 381             /* Compute parameters for interactions between i and j atoms */
 382             qq00             = _mm256_mul_ps(iq0,jq0);
 383
 384             /* EWALD ELECTROSTATICS */
 385
 386             /* Analytical PME correction */
 387             zeta2            = _mm256_mul_ps(beta2,rsq00);
 388             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 389             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 390             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 391             felec            = _mm256_mul_ps(qq00,felec);
 392             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 393             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 394             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 395             velec            = _mm256_mul_ps(qq00,velec);
 396
 397             d                = _mm256_sub_ps(r00,rswitch);
 398             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 399             d2               = _mm256_mul_ps(d,d);
 400             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 401
 402             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 403
 404             /* Evaluate switch function */
 405             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 406             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 407             velec            = _mm256_mul_ps(velec,sw);
 408             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 409
 410             /* Update potential sum for this i atom from the interaction with this j atom. */
 411             velec            = _mm256_and_ps(velec,cutoff_mask);
 412             velec            = _mm256_andnot_ps(dummy_mask,velec);
 413             velecsum         = _mm256_add_ps(velecsum,velec);
 414
 415             fscal            = felec;
 416
 417             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 418
 419             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 420
 421             /* Calculate temporary vectorial force */
 422             tx               = _mm256_mul_ps(fscal,dx00);
 423             ty               = _mm256_mul_ps(fscal,dy00);
 424             tz               = _mm256_mul_ps(fscal,dz00);
 425
 426             /* Update vectorial force */
 427             fix0             = _mm256_add_ps(fix0,tx);
 428             fiy0             = _mm256_add_ps(fiy0,ty);
 429             fiz0             = _mm256_add_ps(fiz0,tz);
 430
 431             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 432             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 433             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 434             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 435             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 436             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 437             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 438             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 439             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 440
 441             }
 442
 443             /* Inner loop uses 109 flops */
 444         }
 445
 446         /* End of innermost loop */
 447
 448         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 449                                                  f+i_coord_offset,fshift+i_shift_offset);
 450
 451         ggid                        = gid[iidx];
 452         /* Update potential energies */
 453         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 454
 455         /* Increment number of inner iterations */
 456         inneriter                  += j_index_end - j_index_start;
 457
 458         /* Outer loop uses 8 flops */
 459     }
 460
 461     /* Increment number of outer iterations */
 462     outeriter        += nri;
 463
 464     /* Update outer/inner flops */
 465
 466     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*109);
 467 }
 468 /*
 469  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_256_single
 470  * Electrostatics interaction: Ewald
 471  * VdW interaction:            None
 472  * Geometry:                   Particle-Particle
 473  * Calculate force/pot:        Force
 474  */
 475 void
 476 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_avx_256_single
 477                     (t_nblist                    * gmx_restrict       nlist,
 478                      rvec                        * gmx_restrict          xx,
 479                      rvec                        * gmx_restrict          ff,
 480                      t_forcerec                  * gmx_restrict          fr,
 481                      t_mdatoms                   * gmx_restrict     mdatoms,
 482                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 483                      t_nrnb                      * gmx_restrict        nrnb)
 484 {
 485     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 486      * just 0 for non-waters.
 487      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 488      * jnr indices corresponding to data put in the four positions in the SIMD register.
 489      */
 490     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 491     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 492     int              jnrA,jnrB,jnrC,jnrD;
 493     int              jnrE,jnrF,jnrG,jnrH;
 494     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 495     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 496     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 497     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 498     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 499     real             rcutoff_scalar;
 500     real             *shiftvec,*fshift,*x,*f;
 501     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 502     real             scratch[4*DIM];
 503     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 504     real *           vdwioffsetptr0;
 505     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 506     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 507     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 508     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 509     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 510     real             *charge;
 511     __m256i          ewitab;
 512     __m128i          ewitab_lo,ewitab_hi;
 513     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 514     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 515     real             *ewtab;
 516     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 517     real             rswitch_scalar,d_scalar;
 518     __m256           dummy_mask,cutoff_mask;
 519     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 520     __m256           one     = _mm256_set1_ps(1.0);
 521     __m256           two     = _mm256_set1_ps(2.0);
 522     x                = xx[0];
 523     f                = ff[0];
 524
 525     nri              = nlist->nri;
 526     iinr             = nlist->iinr;
 527     jindex           = nlist->jindex;
 528     jjnr             = nlist->jjnr;
 529     shiftidx         = nlist->shift;
 530     gid              = nlist->gid;
 531     shiftvec         = fr->shift_vec[0];
 532     fshift           = fr->fshift[0];
 533     facel            = _mm256_set1_ps(fr->epsfac);
 534     charge           = mdatoms->chargeA;
 535
 536     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 537     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff);
 538     beta2            = _mm256_mul_ps(beta,beta);
 539     beta3            = _mm256_mul_ps(beta,beta2);
 540
 541     ewtab            = fr->ic->tabq_coul_FDV0;
 542     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 543     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 544
 545     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 546     rcutoff_scalar   = fr->rcoulomb;
 547     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 548     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 549
 550     rswitch_scalar   = fr->rcoulomb_switch;
 551     rswitch          = _mm256_set1_ps(rswitch_scalar);
 552     /* Setup switch parameters */
 553     d_scalar         = rcutoff_scalar-rswitch_scalar;
 554     d                = _mm256_set1_ps(d_scalar);
 555     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 556     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 557     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 558     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 559     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 560     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 561
 562     /* Avoid stupid compiler warnings */
 563     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 564     j_coord_offsetA = 0;
 565     j_coord_offsetB = 0;
 566     j_coord_offsetC = 0;
 567     j_coord_offsetD = 0;
 568     j_coord_offsetE = 0;
 569     j_coord_offsetF = 0;
 570     j_coord_offsetG = 0;
 571     j_coord_offsetH = 0;
 572
 573     outeriter        = 0;
 574     inneriter        = 0;
 575
 576     for(iidx=0;iidx<4*DIM;iidx++)
 577     {
 578         scratch[iidx] = 0.0;
 579     }
 580
 581     /* Start outer loop over neighborlists */
 582     for(iidx=0; iidx<nri; iidx++)
 583     {
 584         /* Load shift vector for this list */
 585         i_shift_offset   = DIM*shiftidx[iidx];
 586
 587         /* Load limits for loop over neighbors */
 588         j_index_start    = jindex[iidx];
 589         j_index_end      = jindex[iidx+1];
 590
 591         /* Get outer coordinate index */
 592         inr              = iinr[iidx];
 593         i_coord_offset   = DIM*inr;
 594
 595         /* Load i particle coords and add shift vector */
 596         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 597
 598         fix0             = _mm256_setzero_ps();
 599         fiy0             = _mm256_setzero_ps();
 600         fiz0             = _mm256_setzero_ps();
 601
 602         /* Load parameters for i particles */
 603         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 604
 605         /* Start inner kernel loop */
 606         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 607         {
 608
 609             /* Get j neighbor index, and coordinate index */
 610             jnrA             = jjnr[jidx];
 611             jnrB             = jjnr[jidx+1];
 612             jnrC             = jjnr[jidx+2];
 613             jnrD             = jjnr[jidx+3];
 614             jnrE             = jjnr[jidx+4];
 615             jnrF             = jjnr[jidx+5];
 616             jnrG             = jjnr[jidx+6];
 617             jnrH             = jjnr[jidx+7];
 618             j_coord_offsetA  = DIM*jnrA;
 619             j_coord_offsetB  = DIM*jnrB;
 620             j_coord_offsetC  = DIM*jnrC;
 621             j_coord_offsetD  = DIM*jnrD;
 622             j_coord_offsetE  = DIM*jnrE;
 623             j_coord_offsetF  = DIM*jnrF;
 624             j_coord_offsetG  = DIM*jnrG;
 625             j_coord_offsetH  = DIM*jnrH;
 626
 627             /* load j atom coordinates */
 628             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 629                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 630                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 631                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 632                                                  &jx0,&jy0,&jz0);
 633
 634             /* Calculate displacement vector */
 635             dx00             = _mm256_sub_ps(ix0,jx0);
 636             dy00             = _mm256_sub_ps(iy0,jy0);
 637             dz00             = _mm256_sub_ps(iz0,jz0);
 638
 639             /* Calculate squared distance and things based on it */
 640             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 641
 642             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 643
 644             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 645
 646             /* Load parameters for j particles */
 647             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 648                                                                  charge+jnrC+0,charge+jnrD+0,
 649                                                                  charge+jnrE+0,charge+jnrF+0,
 650                                                                  charge+jnrG+0,charge+jnrH+0);
 651
 652             /**************************
 653              * CALCULATE INTERACTIONS *
 654              **************************/
 655
 656             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 657             {
 658
 659             r00              = _mm256_mul_ps(rsq00,rinv00);
 660
 661             /* Compute parameters for interactions between i and j atoms */
 662             qq00             = _mm256_mul_ps(iq0,jq0);
 663
 664             /* EWALD ELECTROSTATICS */
 665
 666             /* Analytical PME correction */
 667             zeta2            = _mm256_mul_ps(beta2,rsq00);
 668             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 669             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 670             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 671             felec            = _mm256_mul_ps(qq00,felec);
 672             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 673             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 674             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 675             velec            = _mm256_mul_ps(qq00,velec);
 676
 677             d                = _mm256_sub_ps(r00,rswitch);
 678             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 679             d2               = _mm256_mul_ps(d,d);
 680             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 681
 682             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 683
 684             /* Evaluate switch function */
 685             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 686             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 687             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 688
 689             fscal            = felec;
 690
 691             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 692
 693             /* Calculate temporary vectorial force */
 694             tx               = _mm256_mul_ps(fscal,dx00);
 695             ty               = _mm256_mul_ps(fscal,dy00);
 696             tz               = _mm256_mul_ps(fscal,dz00);
 697
 698             /* Update vectorial force */
 699             fix0             = _mm256_add_ps(fix0,tx);
 700             fiy0             = _mm256_add_ps(fiy0,ty);
 701             fiz0             = _mm256_add_ps(fiz0,tz);
 702
 703             fjptrA             = f+j_coord_offsetA;
 704             fjptrB             = f+j_coord_offsetB;
 705             fjptrC             = f+j_coord_offsetC;
 706             fjptrD             = f+j_coord_offsetD;
 707             fjptrE             = f+j_coord_offsetE;
 708             fjptrF             = f+j_coord_offsetF;
 709             fjptrG             = f+j_coord_offsetG;
 710             fjptrH             = f+j_coord_offsetH;
 711             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 712
 713             }
 714
 715             /* Inner loop uses 105 flops */
 716         }
 717
 718         if(jidx<j_index_end)
 719         {
 720
 721             /* Get j neighbor index, and coordinate index */
 722             jnrlistA         = jjnr[jidx];
 723             jnrlistB         = jjnr[jidx+1];
 724             jnrlistC         = jjnr[jidx+2];
 725             jnrlistD         = jjnr[jidx+3];
 726             jnrlistE         = jjnr[jidx+4];
 727             jnrlistF         = jjnr[jidx+5];
 728             jnrlistG         = jjnr[jidx+6];
 729             jnrlistH         = jjnr[jidx+7];
 730             /* Sign of each element will be negative for non-real atoms.
 731              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 732              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 733              */
 734             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 735                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 736
 737             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 738             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 739             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 740             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 741             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 742             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 743             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 744             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 745             j_coord_offsetA  = DIM*jnrA;
 746             j_coord_offsetB  = DIM*jnrB;
 747             j_coord_offsetC  = DIM*jnrC;
 748             j_coord_offsetD  = DIM*jnrD;
 749             j_coord_offsetE  = DIM*jnrE;
 750             j_coord_offsetF  = DIM*jnrF;
 751             j_coord_offsetG  = DIM*jnrG;
 752             j_coord_offsetH  = DIM*jnrH;
 753
 754             /* load j atom coordinates */
 755             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 756                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 757                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 758                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 759                                                  &jx0,&jy0,&jz0);
 760
 761             /* Calculate displacement vector */
 762             dx00             = _mm256_sub_ps(ix0,jx0);
 763             dy00             = _mm256_sub_ps(iy0,jy0);
 764             dz00             = _mm256_sub_ps(iz0,jz0);
 765
 766             /* Calculate squared distance and things based on it */
 767             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 768
 769             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 770
 771             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 772
 773             /* Load parameters for j particles */
 774             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 775                                                                  charge+jnrC+0,charge+jnrD+0,
 776                                                                  charge+jnrE+0,charge+jnrF+0,
 777                                                                  charge+jnrG+0,charge+jnrH+0);
 778
 779             /**************************
 780              * CALCULATE INTERACTIONS *
 781              **************************/
 782
 783             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 784             {
 785
 786             r00              = _mm256_mul_ps(rsq00,rinv00);
 787             r00              = _mm256_andnot_ps(dummy_mask,r00);
 788
 789             /* Compute parameters for interactions between i and j atoms */
 790             qq00             = _mm256_mul_ps(iq0,jq0);
 791
 792             /* EWALD ELECTROSTATICS */
 793
 794             /* Analytical PME correction */
 795             zeta2            = _mm256_mul_ps(beta2,rsq00);
 796             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 797             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 798             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 799             felec            = _mm256_mul_ps(qq00,felec);
 800             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 801             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 802             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 803             velec            = _mm256_mul_ps(qq00,velec);
 804
 805             d                = _mm256_sub_ps(r00,rswitch);
 806             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 807             d2               = _mm256_mul_ps(d,d);
 808             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 809
 810             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 811
 812             /* Evaluate switch function */
 813             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 814             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 815             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 816
 817             fscal            = felec;
 818
 819             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 820
 821             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 822
 823             /* Calculate temporary vectorial force */
 824             tx               = _mm256_mul_ps(fscal,dx00);
 825             ty               = _mm256_mul_ps(fscal,dy00);
 826             tz               = _mm256_mul_ps(fscal,dz00);
 827
 828             /* Update vectorial force */
 829             fix0             = _mm256_add_ps(fix0,tx);
 830             fiy0             = _mm256_add_ps(fiy0,ty);
 831             fiz0             = _mm256_add_ps(fiz0,tz);
 832
 833             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 834             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 835             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 836             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 837             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 838             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 839             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 840             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 841             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 842
 843             }
 844
 845             /* Inner loop uses 106 flops */
 846         }
 847
 848         /* End of innermost loop */
 849
 850         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 851                                                  f+i_coord_offset,fshift+i_shift_offset);
 852
 853         /* Increment number of inner iterations */
 854         inneriter                  += j_index_end - j_index_start;
 855
 856         /* Outer loop uses 7 flops */
 857     }
 858
 859     /* Increment number of outer iterations */
 860     outeriter        += nri;
 861
 862     /* Update outer/inner flops */
 863
 864     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*106);
 865 }